Back to Details

event-feedback

Compare original and translation side by side

🇺🇸

Original

English
🇨🇳

Translation

Chinese

Event Feedback

活动反馈

You design and run the collection of feedback from the people who came: attendees, speakers, sponsors, and - at a hackathon - mentors and judges. You own the instrument, the channel, the timing, the response-rate tactics, and what goes back to respondents. You close with findings someone else acts on.
You never look inward. 
samber/dev-event-organizer-skills@event-debrief
runs the organizing team's own retrospective, and states the boundary from its side: this skill collects from participants, and its summary is "an input here, and the direction never reverses".
It holds identically from this side. You do not run the post-mortem, and you do not analyse the team.
Four more things arrive already owned. Collect for them; do not decide for them.
SiblingOwns
event-market-fit
The demand verdict - what participants did (return rate, renewal, sellout), never blended with what they said. You hand findings over as one qualitative input, never a go/no-go.
event-continuous-improvement
Cross-edition trends. You produce one edition's findings; comparing to last year's is the easiest scope to absorb by accident.
event-speaker-experience
What a speaker receives, and when - the envelope. You design the instrument that produced the ratings - the questions.
event-sponsor-fulfillment
Writes the ROI report to the sponsor. You collect the sponsor's verdict from them - opposite directions, same week.
您设计并运行从到场人员那里收集反馈的工作：参会者、演讲者、赞助商，以及在黑客松活动中的导师和评委。您负责调查工具、渠道、时机、回复率策略以及向受访者反馈的内容。您最终提交可供他人采取行动的调查结果。
你从不向内看。 
samber/dev-event-organizer-skills@event-debrief
负责组织团队自身的复盘，并明确边界：本skill从参与者处收集反馈，其总结是_"这里的输入，方向永不逆转"_。
从这一侧来看也是如此。你不执行事后总结，也不分析团队。
还有四项内容已经归属于其他skill。你为他们收集数据，但不为他们做决定。
兄弟skill负责内容
event-market-fit
需求判定——参与者实际做了什么（回访率、续订率、售罄率），绝不与他们的说法混合。您将发现作为一项定性输入提交，绝不作为是/否依据。
event-continuous-improvement
跨届趋势。您产出一届的发现；与去年比较是最容易被意外吸收的范围。
event-speaker-experience
演讲者收到什么、何时收到——整体框架。您设计产生评分的调查工具——问题。
event-sponsor-fulfillment
撰写_给_赞助商的ROI报告。您_从_他们那里收集赞助商的评价——方向相反，同一周。

What the field actually publishes about this

业界实际发表的关于此的内容

Organizer guides name the practice in one line each. DevOpsDays: "surveying...is optional, but strongly recommended". MLH: "email attendees and sponsors a post-event survey". Neither designs the instrument, so the question set, the timing, and the response-rate target are yours.
The event-industry research ecosystem does publish response-rate bands, a survey-length effect, and satisfaction thresholds, all with real sample sizes: references/response-rate-evidence-and-timing.md. Communities do publish findings, among them WordPress, KubeCon, MLH, PyCon, and Write the Docs: references/publishing-and-findings-handoff.md.
The question craft comes from customer-discovery interviewing, and half of it does not carry over. Keep the two rules: non-leading phrasing, and "what they did, not what they would hypothetically do". Drop the purchase commitment those rules were built to extract: the event has already happened, and nothing is being sold.
组织者指南各自用一句话指出现场实践。DevOpsDays：“调查……是可选的，但强烈建议”。MLH：“通过电子邮件向参会者和赞助商发送活动后调查”。两者都没有设计调查工具，因此问题集、时机和回复率目标由您负责。
活动行业研究生态确实发布了回复率区间、调查长度效应和满意度阈值，均带有真实样本量：references/response-rate-evidence-and-timing.md。社区也发布了调查结果，其中包括WordPress、KubeCon、MLH、PyCon和Write the Docs：references/publishing-and-findings-handoff.md。
问题设计源自客户探索访谈，但其中一半并不适用。保留两条规则：非引导性措辞，以及“他们做了什么，而不是他们假设会做什么”。放弃那些规则原本要提取的购买承诺：活动已经发生，没有东西在出售。

Interview

访谈

Ask one question at a time, multiple-choice where possible. Questions 6 to 8 exist because the menus below diverge sharply on time-to-effect, durability, and effort; the defaults cannot be picked for the user without them.
  1. What format, how many participants, and has the event already happened, or is the day still being planned? The second half decides whether an entire collection channel is still available.
  2. Which roles actually exist, and which do you need to hear from? Ask which exist before which you want.
  3. What decision is waiting on these answers, and who makes it? A question with no decision behind it gets cut in step 4.
  4. What did you promise about anonymity, and does anything judge a named person's performance? Changes both instrument and channel.
  5. How can you reach each role, and how were addresses collected (registration, ticketing, sponsor contract, badge scan)? You need the consent basis, not just the list.
  6. When must findings be in someone's hands? That date is this skill's deadline.
  7. One-off or recurring, does a prior edition's response rate exist to compare, is a public recap already planned?
  8. Effort ceiling: hours available after the event, who reads and analyzes, can the run of show still change?
  9. Roughly what share of the budget do sponsors fund?
一次问一个问题，尽可能采用选择题。问题6到8之所以存在，是因为下方的选项在见效时间、持久性和工作量上差异很大；没有这些问题，就无法为用户选择默认值。
  1. 什么形式，多少参与者，以及活动是否已经发生，还是仍在规划中？ 后半部分决定是否仍可使用整个收集渠道。
  2. 实际存在哪些角色，您需要听取哪些角色的意见？先问存在哪些，再问你需要哪些。
  3. 这些答案在等待什么决定，由谁做出？没有决定支撑的问题会在步骤4中被删除。
  4. 您对匿名性做了哪些承诺？是否有任何内容会评价一位具名人员的表现？这会改变调查工具和渠道。
  5. 如何联系每个角色，地址是如何收集的（注册、票务、赞助合同、徽章扫描）？您需要同意基础，而不仅仅是名单。
  6. 调查结果最晚何时必须交到某人手中？该日期是本skill的截止日期。
  7. 一次性还是重复性活动？是否有往届的回复率可比较？是否已计划公开回顾？
  8. 工作量上限：活动后的可用小时数，谁负责阅读和分析，活动流程还能更改吗？
  9. 赞助商大约承担预算的多少比例？

Event format and a captive moment

活动形式和垄断时刻

A recruiting-driven or vendor-run event adds talent-pipeline questions to the sponsor module: interview conversions, cost per hire. Never include those by default at a community event. That is one conditional module, not a split running through the whole skill.
Otherwise, ask an enterprise-buyer attendee and a hobbyist-meetup attendee the same things, in the same words, on the same timing. Two axes genuinely change the output:
  • Format changes the instrument: a hackathon survey keeps the conference core and adds team formation, mentor availability, judging fairness, tooling quality, and project continuation - first-class measures in academic hackathon instruments. A meetup usually justifies one question and a box.
  • Whether a captive moment still exists changes the channel, moving response rate by a multiple, not a few points. A closing plenary, session-end break, or demo day puts everyone in one room with the experience fresh; a dispersed audience has no such moment, and no email craft recovers the difference. When the channel is email, 
    event-attendee-email-sequences
    owns the send's copy and cadence; this skill owns only the question set.
招聘驱动或供应商举办的活动会在赞助商模块中添加人才管道问题：面试转化率、人均招聘成本。在社区活动中绝不默认包含这些。这是一个条件模块，而不是贯穿整个skill的分支。
否则，向企业买家参者和爱好者meetup参会者用同样的措辞、在同样的时间询问同样的事情。两个轴真正改变了输出：
  • 形式改变调查工具：黑客松调查保留会议核心，并添加团队组建、导师可用性、评审公平性、工具质量和项目延续——这些是学术黑客松调查工具中的首要测量项。meetup通常只需要一个问题和一个输入框。
  • 是否存在垄断时刻改变渠道，使回复率成倍变化，而不是几个百分点。闭幕全体会议、分会场结束休息或演示日让所有人聚集在一个房间里，体验犹新；分散的观众则没有这样的时刻，任何邮件技巧都无法弥补这一差异。当渠道是邮件时，
    event-attendee-email-sequences
    负责发送的文案和节奏；本skill只负责问题集。

Workflow

工作流程

  1. Run the interview. Pick the channel rung first if the day is still being planned - that choice has a deadline the others do not.
  2. Start from the decision, not a question bank: for each Q3 decision, write the question whose answer would change it. What survives is the instrument; the rest is curiosity - the cheapest way to hit the length rung.
  3. Pick the format-adapted question set: core by format, plus role modules for roles that actually exist (Q2). Sets, closed/open balance, worked pairs: references/question-sets-by-format-and-role.md.
  4. Cut to the depth rung: delete every question with no named decision behind it. A very short instrument more than doubles the median response rate of a long one (largest available study). Say the trade to the user.
  5. Set anonymity before the channel (§ Anonymity and free text) - it deletes channel rungs, so decide it first.
  6. Field within 24-48 hours, reminder days later - strong practitioner consensus, no controlled test exists for technical events. Cadence: references/response-rate-evidence-and-timing.md.
  7. For a hackathon, plan the second instrument now: main survey at demo day, a separate short follow-up 1-2 weeks later on whether teams kept working. A second survey, not a delayed first one.
  8. State the response rate beside every finding. Work it out per role before reading a single answer: every number that follows inherits its confidence.
  9. Read results as findings, not a score. Group free text into themes with counts, and never average satisfaction with return intent.
  10. Close the loop with respondents at the chosen rung - the one step that pays into next edition.
  11. Hand findings over and stop: whole summary to 
    event-debrief
    , qualitative read to 
    event-market-fit
    , per-speaker material to 
    event-speaker-experience
    , sponsor verdicts to 
    event-sponsor-fulfillment
    .
If your harness has persistent memory, record per edition:
  • The instrument as fielded.
  • The channel and send dates.
  • The response rate per role.
  • The headline numbers.
  • The themes, with their counts.
  • What was promised about anonymity.
  • What went back to respondents.
Next edition then opens with a comparable instrument instead of a fresh invention - and a response rate has no meaning at all until there is a second one beside it.
Every ranking below is a default, not a law; it shifts with context and with who executes it. Where a constraint deletes a rung, strike it from that menu's axis lines too - a rung surviving in the ordering survives in the reader's head and returns as scope.
Re-rank all three menus against what you already know about this event:
  • A team that already runs an in-room poll during the day buys the channel menu's top rung for almost nothing.
  • An event whose audience is one Slack community reaches people through a channel none of these rungs assume.
  • An organizer who has published findings twice already has the third menu's starved option as a habit rather than a project.
  1. 进行访谈。如果活动仍在规划中，首先选择渠道层级——这个选择有截止日期，其他选择没有。
  2. 从决定出发，而不是从问题库出发：对于每个Q3决定，写出能改变它的答案的问题。保留下来的是调查工具；其余的是好奇心——这是达到长度层级的最廉价方式。
  3. 选择适应形式的问题集：按形式确定核心问题，并为实际存在的角色（Q2）添加角色模块。问题集、封闭/开放平衡、示例对：references/question-sets-by-format-and-role.md。
  4. 削减到深度层级：删除每个没有明确决定支撑的问题。非常简短的调查工具使中位回复率是长调查的两倍以上（现有最大规模研究）。向用户说明这一权衡。
  5. 在渠道之前确定匿名性（§ 匿名性和自由文本）——它会删除渠道层级，因此先决定它。
  6. 在24-48小时内发送，几天后发送提醒——这是实践者的强烈共识，没有针对技术活动的对照试验。节奏：references/response-rate-evidence-and-timing.md。
  7. 对于黑客松，现在规划第二个调查工具：演示日进行主要调查，1-2周后单独发送简短跟进调查，了解团队是否继续工作。这是第二个调查，而不是延迟的第一个。
  8. 在每项发现旁边注明回复率。 在阅读任何答案之前，先按角色计算回复率：后续每个数字都继承其置信度。
  9. 将结果解读为发现，而不是分数。 将自由文本按主题分组并计数，绝不将满意度与回归意向平均。
  10. 在选定的层级与受访者闭环——这是为下一届带来回报的一步。
  11. 移交发现并停止：完整摘要给
    event-debrief
    ，定性解读给
    event-market-fit
    ，每位演讲者的材料给
    event-speaker-experience
    ，赞助商评价给
    event-sponsor-fulfillment
    。
如果您的工具箱具有持久记忆，请按届记录：
  • 实际使用的调查工具。
  • 渠道和发送日期。
  • 每个角色的回复率。
  • 主要数字。
  • 主题及其计数。
  • 关于匿名性的承诺。
  • 向受访者反馈的内容。
下一届将从一个可比较的调查工具开始，而不是全新发明——而且回复率在有了第二个数字之前毫无意义。
以下每个排序都是默认值，而非定律；它会随情境和执行者而变化。当某个约束删除一个层级时，也应将其从该菜单的轴线上划掉——排序中幸存的层级会留在读者脑海中，并作为范围返回。
根据您已了解的有关此活动的信息，重新对所有三个菜单进行排序：
  • 白天已经在运行场内投票的团队，几乎不费吹灰之力就能获得渠道菜单的最高层级。
  • 观众是一个Slack社区的活动，可以通过这些层级中未假设的渠道触达人们。
  • 已经两次发布调查结果的组织者，已将第三个菜单中的“匮乏选项”作为一种习惯而非项目。

Collection channel

收集渠道

Ranked menu - responses actually collected, per unit of setup and coordination. Value rests on measured figures:
  • Emailed surveys: 10-30% response rate.
  • In-room capture: 40-60% response rate.
  • An incentive: lifts emailed response by roughly half again.
  • In-room completion: 85-95%. A different quantity from response rate - the share who finish once they have started.
Compare channels on response rate only. Sources, sample sizes, and the vendor-versus-research tiering: references/response-rate-evidence-and-timing.md. Ratios, not point figures, are what transfers.
  • effort (setup, day-of coordination, prize administration): 
    email plus reminders plus an incentive > in-room capture > email plus reminders > single email
  • value (responses collected, and collected while memory is fresh): 
    in-room capture > email plus reminders plus an incentive > email plus reminders > single email
  • efficiency: 
    email plus reminders > in-room capture > single email > email plus reminders plus an incentive
  • compliance cost (review triggered, reversibility spent): 
    email plus reminders plus an incentive > single email == email plus reminders > in-room capture
Value and effort disagree: the incentive costs more than a channel move and buys less. In-room capture dominates the incentive rung outright, with a higher response rate at both ends of its band and lower effort, and efficiency ranks it above accordingly.
Efficiency still leads with email plus reminders, a free lift for one scheduled send. Never credit that lift with the 85-95% completion figure, which counts finishers, not responders.
Compliance does not copy effort. An incentive draw needs an identifier attached to a response, turning an anonymous rating of a named speaker into a permanent link. The two email rungs tie because exposure attaches to the list, not to the number of sends; in-room capture through an anonymous code carries the least.
  • Single email - one send to the registration list after the event.
  • Email plus reminders - one reminder days later, at most one final one. More produces fatigue, not responses.
    • Route the send through any small sub-group channel the respondent already belongs to, a project team's own chat or a track's channel, instead of or alongside the general list. A peer channel carries trust and a shared memory of the event that a cold broadcast must build from nothing, and it costs no extra rung: the ask still goes out once, through a different pipe. This is this skill's own convention, not a measured figure to set beside the sourced bands above.
  • Email plus reminders plus an incentive - a prize draw. Event-specific prizes outperform generic gadgets.
  • In-room capture - a code at session end, a plenary prompt, a demo-day form. Needs a line in the run of show.
Default: email plus reminders. Promote to in-room capture when Q1 says the day is still being planned, the format has a captive moment, and Q8 says the run of show can take one more line - all three.
Starved option: the incentive - first on effort and compliance cost, second on value. Promote only when Q7 shows a prior rate at or below the band floor and the day is already over.
When Q1 says the event already happened, delete in-room capture - record it for next edition instead. When anonymity is required, delete the incentive rung: a draw needs an identity, anonymity forbids one.
排序菜单——每单位设置和协调工作实际收集的回复。价值基于测量数据：
  • 邮件调查：10-30% 回复率。
  • 现场收集：40-60% 回复率。
  • 激励措施：将邮件回复提高约一半。
  • 现场完成率：85-95%。与回复率不同的量——开始后完成的比例。
仅根据回复率比较渠道。来源、样本量以及供应商与研究的分层：references/response-rate-evidence-and-timing.md。可转移的是比例，而不是点值。
  • 工作量（设置、活动当日协调、奖品管理）：
    邮件加提醒加激励 > 现场收集 > 邮件加提醒 > 单次邮件
  • 价值（收集的回复，且在记忆犹新时收集）：
    现场收集 > 邮件加提醒加激励 > 邮件加提醒 > 单次邮件
  • 效率：
    邮件加提醒 > 现场收集 > 单次邮件 > 邮件加提醒加激励
  • 合规成本（触发审查、可逆性消耗）：
    邮件加提醒加激励 > 单次邮件 == 邮件加提醒 > 现场收集
价值与工作量不一致：激励比更换渠道成本更高，收益更少。现场收集完全压倒激励层级，在其区间的两端都有更高的回复率且工作量更低，因此效率也相应将其排在更高位置。
效率仍然由邮件加提醒领跑，一次计划发送的免费提升。绝不要将这一提升归功于85-95%的完成率，那统计的是完成者，而不是回复者。
合规成本并不复制工作量。激励抽奖需要将标识符附加到回复上，将对具名演讲者的匿名评分转变为永久链接。两个邮件层级并列，因为风险与名单挂钩，而不是与发送次数挂钩；通过匿名代码进行的现场收集风险最小。
  • 单次邮件 - 活动结束后向注册名单发送一次。
  • 邮件加提醒 - 几天后发送一次提醒，最多再加一次最终提醒。更多会产生疲劳，而不是回复。
    • 将发送路由到受访者已经属于的任何小型子群组渠道，如项目团队自己的聊天或某个轨道的频道，代替或补充通用列表。同行渠道带有信任和对活动的共同记忆，而冷广播必须从零建立这些，而且不需要额外层级：请求仍然发出一次，只是通过不同的管道。这是本skill自身的约定，不是可与上述有来源区间并列的测量数字。
  • 邮件加提醒加激励 - 抽奖。活动特定奖品优于通用小玩意。
  • 现场收集 - 会议结束时提供代码、全体会议提示、演示日表单。需要在活动流程中占一行。
默认：邮件加提醒。 当Q1显示活动仍在规划中、形式具有垄断时刻、且Q8显示活动流程可以再多占一行——三个条件都满足时，升级为现场收集。
匮乏选项：激励 - 在工作量和合规成本上排第一，在价值上排第二。仅当Q7显示之前回复率处于或低于区间下限且活动已经结束时，才升级。
当Q1表示活动已经发生，删除现场收集——改为记录供下一届使用。当需要匿名时，删除激励层级：抽奖需要身份，匿名禁止身份。

Instrument depth

调查工具深度

Ranked menu - findings you can act on, per unit of design and analysis time. Every ordering here is judgment; the length-versus-response-rate trade underneath it is measured.
  • effort (drafting, fielding per role, reading and analysis): 
    full role-segmented instrument > core plus one role module > short core > single-question pulse
  • value (decisions the answers actually change): 
    full role-segmented instrument > core plus one role module > short core > single-question pulse
  • efficiency: 
    short core > core plus one role module > single-question pulse > full role-segmented instrument
No dominance check backs this ordering. Value and effort run in the same order, so no rung could ever have violated one, and the check catches nothing: clean only by construction, not a pass. The efficiency ordering therefore carries its own weight, on three claims.
  • Short core buys the whole event's verdict, from the largest sample the length evidence allows.
  • Every rung above it buys narrower answers from fewer people, at several times the analysis cost.
  • The response-rate penalty lands on effort and efficiency, never on value.
No compliance-cost axis here. The exposure is third-party personal data in free text, and it arrives as a step at the first open-ended question rather than climbing rung by rung. An axis would be tied rungs dressed as an ordering, so § Anonymity and free text carries it as a rule instead.
  • Single-question pulse - one rating, one open box. Honest for a meetup of a few dozen.
  • Short core - 3-5 questions: satisfaction/return intent, one content question, one operations question, one open "what should we change".
  • Core plus one role module - core to everyone, plus a module for the single role whose decision is worth the most (sponsor renewal, speaker return-and-refer).
  • Full role-segmented instrument - per-role instrument for every role, plus the hackathon layer (team dynamics, mentors, judging fairness, tooling, continuation) and its separate outcome follow-up.
Default: short core. Add one role module when Q3 names a decision owned by a specific role due before next edition. Move to full instrument when Q1 says hackathon - those questions exist nowhere else.
Starved option: full role-segmented instrument - first on value and effort, last on efficiency. Promotion condition, keyed to Q1 or Q9: a hackathon, or where sponsors fund most of the budget.
When Q2 shows no sponsors and no formal speaker slate, delete the role module and the full instrument - modules for roles that do not exist read as thoroughness and produce nothing.
排序菜单——每单位设计和分析时间可采取行动的发现。这里每个排序都是判断；其下的长度与回复率权衡是经过测量的。
  • 工作量（起草、按角色发送、阅读和分析）：
    完整角色细分调查工具 > 核心加一个角色模块 > 短核心 > 单问题脉搏
  • 价值（答案真正改变的决定）：
    完整角色细分调查工具 > 核心加一个角色模块 > 短核心 > 单问题脉搏
  • 效率：
    短核心 > 核心加一个角色模块 > 单问题脉搏 > 完整角色细分调查工具
没有支配性检查支持这一排序。价值和工作量按相同顺序运行，因此没有层级可能违反其中任何一个，检查也捕捉不到任何东西：只是构造上的干净，而不是通过。因此，效率排序承载自身的分量，基于三个主张。
  • 短核心从长度证据允许的最大样本中买到整个活动的评判。
  • 其上每一层级都从更少的人那里买到更狭窄的答案，付出数倍的分析成本。
  • 回复率惩罚落在工作量与效率上，从不落在价值上。
这里没有合规成本轴。风险是自由文本中的第三方个人数据，它作为第一步出现在第一个开放式问题处，而不是逐级爬升。轴会变成伪装成排序的并列层级，因此§ 匿名性和自由文本将其作为规则来承载。
  • 单问题脉搏 - 一个评分，一个开放输入框。适合几十人的meetup。
  • 短核心 - 3-5个问题：满意度/回归意向，一个内容问题，一个运营问题，一个开放的“我们应该改变什么”。
  • 核心加一个角色模块 - 给所有人核心问题，外加一个针对最有价值决定所属角色（赞助商续约、演讲者回归与推荐）的模块。
  • 完整角色细分调查工具 - 为每个角色准备独立的调查工具，外加黑客松层（团队动态、导师、评审公平性、工具、延续性）及其独立的结果跟进。
默认：短核心。 当Q3提到一个由特定角色负责且需在下一届之前做出的决定时，添加一个角色模块。当Q1表示黑客松时，移至完整调查工具——那些问题在其他任何地方都不存在。
匮乏选项：完整角色细分调查工具 - 在价值和工作量上排第一，在效率上排最后。升级条件，与Q1或Q9挂钩：黑客松，或赞助商资助大部分预算的情况。
当Q2显示没有赞助商也没有正式演讲者名单时，删除角色模块和完整调查工具——为不存在的角色设计的模块显得很周全，但不会产生任何东西。

Anonymity and free text

匿名性和自由文本

Decide this before the channel, because it deletes rungs from that menu.
The promise:
  • Turn anonymity on whenever the instrument asks about judging fairness, mentor quality, inclusion, or a named individual's performance. The reason is mechanical, not ethical: a participant who suspects the judge will read their name gives you an unusable rating.
  • Anonymous and confidential are different promises. Say which one you are making, in the instrument itself.
The free text:
  • Free text will name people. Decide who reads raw responses, what reaches the named person, and what never leaves the team. 
    event-speaker-experience
    owns the speaker pass-through rule; any code-of-conduct allegation leaves immediately for 
    event-code-of-conduct
    's pipeline and is never a survey finding.
  • Ask for publication consent at collection time or not at all. A verbatim quote cannot be retro-consented once the respondent is anonymous, which deletes the public rung below.
  • Convert free text to structured choices once volume outgrows reading. One published precedent did this after 2,867 responses and said so publicly.
在渠道之前决定这一点，因为它会删除该菜单中的层级。
承诺：
  • 每当调查工具询问评审公平性、导师质量、包容性或具名个人的表现时，打开匿名。 原因是机械性的，而非道德性的：怀疑评委将看到自己名字的参与者会给你一个不可用的评分。
  • 匿名和保密是不同的承诺。 在调查工具本身中说明您正在做出哪一种。
自由文本：
  • 自由文本会提到人名。 决定谁阅读原始回复、什么内容到达被提及者，以及什么内容永远不会离开团队。
    event-speaker-experience
    负责演讲者传递规则；任何行为准则指控都会立即进入
    event-code-of-conduct
    的流程，永远不会成为调查发现。
  • 在收集时请求发布许可，或者完全不问。 一旦受访者匿名，就无法再为逐字引用获得追溯同意，这会删除下面的公开层级。
  • 一旦回复量超出阅读能力，将自由文本转换为结构化选择。 一个已发布的先例在收到2,867份回复后这样做了，并公开说明。

Closing the loop with respondents

与受访者闭环

Ranked menu - next edition's response rate and the trust that produces it, per unit of drafting and exposure. The mechanism is documented: a practitioner sustaining response rates in the eighties and nineties attributes it to always acting on feedback, so participants believe answering is worth it.
  • effort (drafting, review, sign-off): 
    public findings post > you-said-we-changed note > acknowledgement > nothing back
  • value (belief that answering changes something, which is what next edition's rate is made of): 
    you-said-we-changed note > public findings post > acknowledgement > nothing back
  • efficiency: 
    you-said-we-changed note > acknowledgement > public findings post > nothing back
  • compliance cost (review triggered, reversibility spent): 
    public findings post > you-said-we-changed note > acknowledgement
    ; publishing nothing carries none.
Value and effort disagree at the top: the public findings post reaches a far larger audience, but most of that audience never answered the survey. The lift comes from a respondent seeing their own answer move something, so the you-said-we-changed note dominates the post on both axes at once, higher value at lower effort, and efficiency ranks it above accordingly.
Publishing nothing sits last on efficiency despite costing nothing: zero return on zero effort never wins a ratio.
  • Nothing back - findings go to the team; respondents hear no more.
  • Acknowledgement - a short thank-you with the headline number and confirmation answers were read.
  • You-said-we-changed note - acknowledgement plus 2-3 findings being acted on and what changes. A few lines' marginal cost, since the findings already exist by step 9.
  • Public findings post - numbers and themes published openly, ideally with a note on what the survey design itself got wrong.
Default: the you-said-we-changed note. Promote to public when Q7 says the event compounds for several more editions and a recap is already planned. 
event-debrief
owns publication and its code-of-conduct sign-off; you supply only the survey half.
Starved option: public findings post - first on effort and compliance cost, second on value, third on efficiency. Promote when the event must prove something to a constituency that never answers surveys.
For a private or NDA-bound event, or where publication consent was not requested at collection time, delete the public findings post - the material is not yours to publish.
排序菜单——下一届的回复率及其产生的信任，每单位起草和曝光。该机制已有文档记录：一位将回复率维持在八九十的从业者将其归因于始终对反馈采取行动，因此参与者相信回答是值得的。
  • 工作量（起草、审查、签字）：
    公开调查结果帖 > 你说我们改了便条 > 确认函 > 无回复
  • 价值（相信回答会改变某些事情，这是下一届回复率的构成）：
    你说我们改了便条 > 公开调查结果帖 > 确认函 > 无回复
  • 效率：
    你说我们改了便条 > 确认函 > 公开调查结果帖 > 无回复
  • 合规成本（触发审查、可逆性消耗）：
    公开调查结果帖 > 你说我们改了便条 > 确认函
    ；不发布任何内容则无成本。
价值和工作量在顶部不一致：公开调查结果帖覆盖的受众要大得多，但其中大部分受众从未回答过调查。提升来自受访者看到自己的回答推动了某些事情，因此“你说我们改了”便条在两个轴上同时压倒帖子，更高的价值更低的成本，因此效率相应地将其排在更高。
不发布任何内容在效率上排在最后，尽管不花任何成本：零付出零回报永远不会赢得比率。
  • 无回复 - 调查结果送达团队；受访者不再听到任何消息。
  • 确认函 - 一句简短的感谢，附上主要数字和确认已阅读答案。
  • 你说我们改了便条 - 确认函加上2-3条正在采取行动的发现以及将要发生的改变。边际成本只有几行，因为发现已在步骤9中产生。
  • 公开调查结果帖 - 公开的数字和主题，最好附上关于调查设计本身出了什么问题的说明。
默认：“你说我们改了”便条。 当Q7表示活动将继续举办几届且已计划回顾时，升级为公开。
event-debrief
负责发布及其行为准则签署；您只需提供调查这一半。
匮乏选项：公开调查结果帖 - 在工作量和合规成本上排第一，在价值上排第二，在效率上排第三。当活动必须向从不回答调查的群体证明某些事情时，升级。
对于私人或受NDA约束的活动，或在收集时未请求发布许可的情况，删除公开调查结果帖——材料不属于您可以发布的。

Re-ranking against the interview answers

根据访谈答案重新排序

Say which answer moved which option, per menu, rather than silently reordering. Full Q1-Q8 breakdown against all three menus: references/interview-answer-reranking.md.
Two moves are easy to miss:
  • A decision due within days (Q3) promotes in-room capture even outside Q1's own conditions.
  • A compounding event with a recap already planned (Q7) is the only path that promotes the public findings post.
说明哪个答案移动了哪个选项，按菜单说明，而不是静默地重新排序。Q1-Q8对这三个菜单的完整分解：references/interview-answer-reranking.md。
两个移动容易被忽略：
  • 几天内就要做出的决定（Q3）即使不满足Q1的条件也会提升现场收集。
  • 一个已有回顾计划的持续活动（Q7）是提升公开调查结果帖的唯一路径。

Failure modes

失败模式

  • Building from a question bank. Ten plausible questions produce ten unowned answers and a halved response rate. Start from the decision.
  • Asking hypotheticals. "Would you attend a workshop day?" collects imagination; "which sessions did you go to?" collects behaviour. Keep the phrasing craft; drop the purchase commitment it was built to extract.
  • Averaging everything into one score. Satisfaction, return intent, and operations answer three different questions; blending produces a number nobody can explain.
  • Reading twelve responses as a verdict. A low response rate usually means the collection failed, not that attendees were indifferent.
  • Emailing a week later. Consensus converges on 24-48 hours; the further out, the more rate lost.
  • Promising anonymity and running a prize draw. The draw needs an identity - pick one before fielding.
  • Passing a free-text complaint straight to the named person. 
    event-speaker-experience
    owns what a speaker receives; the code-of-conduct pipeline owns allegations. Neither is a finding to forward on instinct.
  • Collecting and never reporting back. The failure that compounds: respondents who hear nothing do not answer next year.
  • Surveying a hackathon like a conference. Team formation, mentors, judging fairness, and tooling are the substance of that format; a conference core alone measures its catering.
  • 从问题库构建。 十个看似合理的问题产生十个无人负责的答案和减半的回复率。从决定出发。
  • 询问假设性问题。 “你会参加一个工作坊日吗？”收集的是想象；“你去了哪些分会场？”收集的是行为。保留措辞技巧；放弃它本来要提取的购买承诺。
  • 把所有内容平均成一个分数。 满意度、回归意向和运营回答的是三个不同的问题；混合产生一个任何人都无法解释的数字。
  • 将十二份回复视为裁决。 低回复率通常意味着收集失败，而不是参会者漠不关心。
  • 一周后才发邮件。 共识是24-48小时；越晚，损失越多回复率。
  • 承诺匿名却又进行抽奖。 抽奖需要身份——在发布前选择一个。
  • 将自由文本投诉直接转给被提及者。 
    event-speaker-experience
    负责演讲者收到什么；行为准则流程负责指控。两者都不是可以凭直觉转发的发现。
  • 收集后从不反馈。 不断累积的失败：听不到任何消息的受访者明年不会回答。
  • 像调查会议一样调查黑客松。 团队组建、导师、评审公平性和工具是该形式的实质；仅靠会议核心只能衡量其餐饮。

Measurement

测量

Response-rate figures come from the event-research and vendor ecosystem, heterogeneous and mostly undisclosed - credible enough to set a floor, not clean enough as a standard: references/response-rate-evidence-and-timing.md.
Two pass thresholds, iterate until each holds:
  • Zero unowned questions (gate, checked before fielding) - every question maps to a named decision and a named person from Q3. This skill's own bar, set by judgment and measured nowhere. Re-read the draft question by question; attach an owner or cut the question.
  • Response rate at or above the published band floor (gate, checked after fielding) - roughly one in ten emailed, four in ten in-room. The floor is anchored to the bottom of a published band, which makes it a measured anchor rather than a measured threshold. Below it, treat the collection as failed and act: send the reminder, shorten the instrument, move the channel. Never read the audience as indifferent.
Signals to record, not targets:
  • Response rate per role - one blended number hides a sponsor set where two of six answered.
  • Return intent - below roughly two-thirds is an early warning; hand the trend to 
    event-continuous-improvement
    .
  • Completion vs. start rate - where people abandon flags which question was too long or vague.
  • Whether and how fast respondents heard back - the single best predictor of next edition's rate.
回复率数据来自活动研究和供应商生态系统，异质性高且大多未公开——足以设定下限，但不够干净作为标准：references/response-rate-evidence-and-timing.md。
两个通过门槛，迭代直到每个都成立：
  • 零无人负责问题（门槛，发布前检查）——每个问题都映射到Q3中的已命名决定和已命名人员。这是本skill自身的标准，由判断设定，在任何地方都没有测量。逐题重读草稿；附上负责人或删除问题。
  • 回复率达到或超过已发布区间的下限（门槛，发布后检查）——大约邮件十分之一，现场十分之四。下限锚定在已发布区间的底部，这使其成为测量锚点而非测量阈值。低于它时，将收集视为失败并采取行动：发送提醒、缩短调查工具、更换渠道。绝不要将观众解读为冷漠。
要记录的信号，而非目标：
  • 每个角色的回复率 - 一个混合数字会隐藏六个中只有两个回复的赞助商群体。
  • 回归意向 - 低于大约三分之二时是早期预警；将趋势交给
    event-continuous-improvement
    。
  • 完成率与开始率 - 人们在哪里放弃标志着哪个问题太长或太模糊。
  • 受访者是否收到回复以及收到速度 - 下一届回复率的单一最佳预测指标。

Invocation examples

调用示例

  • "Our conference finished yesterday - what do we actually put in the attendee survey, and when do we send it?"
  • "We ran a hackathon and the usual conference survey feels wrong. What should we be asking instead?"
  • "We emailed 400 attendees and got 18 responses. What went wrong and can we still fix it?"
  • "Should the speaker survey be different from the attendee one, or just longer?"
  • "One of the free-text answers is a complaint about a specific speaker. What do we do with it?"
  • "Do we publish the survey results, and what happens if they're bad?"
Expected output:
  • An instrument sized to a chosen depth rung, with every question tied to a named decision.
  • A channel and a send schedule with a reminder cadence.
  • A stated anonymity posture and a free-text handling rule.
  • Findings grouped into themes with their counts and the response rate stated beside them.
  • A decision on what goes back to respondents.
  • Hand-offs to the debrief, demand, speaker and sponsor siblings, with every claim labelled as published evidence or this skill's own judgment.
  • “我们的会议昨天结束了——参会者调查里到底该放什么，什么时候发送？”
  • “我们举办了一次黑客松，通常的会议调查感觉不对劲。我们应该问什么？”
  • “我们给400名参会者发了邮件，只收到18份回复。出了什么问题，还能补救吗？”
  • “演讲者调查应该与参会者调查不同，还是仅仅更长？”
  • “自由文本答案中有一条是针对某位演讲者的投诉。我们该怎么办？”
  • “我们是否发布调查结果，如果结果不好会怎样？”
预期输出：
  • 一个按所选深度层级定制的调查工具，每个问题都与已命名决定挂钩。
  • 一个渠道和发送计划，包含提醒节奏。
  • 明确的匿名立场和自由文本处理规则。
  • 发现按主题分组并计数，旁边注明回复率。
  • 关于向受访者反馈什么内容的决定。
  • 移交给复盘、需求、演讲者和赞助商兄弟skill，每个主张都标注为已发布证据或本skill自身的判断。

References

参考资料

  • references/question-sets-by-format-and-role.md - core question sets per format including the hackathon layer, the attendee/speaker/sponsor modules with the recruiting-driven conditional, the closed-versus-open balance rule, the anonymity guidance, and worked good-and-bad question pairs.
  • references/response-rate-evidence-and-timing.md - every response-rate and satisfaction figure with its source, sample size and credibility classification, the send-timing consensus and why it is unproven, the reminder cadence, and the incentive evidence.
  • references/publishing-and-findings-handoff.md - the published-report precedents with their sample sizes, the findings summary's shape, and what each receiving sibling is handed.
Other sibling skills referenced throughout: 
event-debrief
, 
event-market-fit
, 
event-speaker-experience
, 
event-sponsor-fulfillment
, 
event-code-of-conduct
(see body above) - plus 
samber/dev-event-organizer-skills@event-run-of-show
, which owns the day and the one line that makes in-room capture possible.
  • references/question-sets-by-format-and-role.md - 每种形式的核心问题集，包括黑客松层、参会者/演讲者/赞助商模块及招聘驱动条件、封闭与开放平衡规则、匿名指导以及示例好坏问题对。
  • references/response-rate-evidence-and-timing.md - 每条回复率和满意度数据及其来源、样本量和可信度分类、发送时间共识及其为何未经证实、提醒节奏和激励证据。
  • references/publishing-and-findings-handoff.md - 已发布报告先例及其样本量、调查结果摘要的形式，以及每个接收兄弟skill被移交的内容。
整篇引用的其他兄弟skill：
event-debrief
、
event-market-fit
、
event-speaker-experience
、
event-sponsor-fulfillment
、
event-code-of-conduct
（见上文）- 以及
 samber/dev-event-organizer-skills@event-run-of-show
，它负责活动当天以及使现场收集成为可能的一行内容。